This is for building voice AI agents on LiveKit Cloud with an opinionated approach that treats low latency as non-negotiable. It pushes you toward LiveKit Inference to avoid juggling separate API keys, mandates test coverage for everything you ship, and structures complex agents using handoffs between specialized agents rather than one bloated monolith. The skill is adamant about never trusting LLM memory for API details and requires either using the LiveKit MCP server or explicitly marking unverified code. If you're self-hosting LiveKit instead of using their cloud, some guidance won't apply. The philosophy here is that voice conversations die from context bloat and slow responses, so it architects around keeping prompts lean and tools minimal.
npx skills add https://github.com/livekit/agent-skills --skill livekit-agents